DeepMind Maps Global Forest Typology at 10-Metre Resolution

Google DeepMind has introduced ForTy, a machine learning model that classifies global forest structures at a 10-metre resolution.

Existing satellite monitoring systems generally measure canopy cover alone. By assessing only whether trees exist above a specific height threshold, these datasets frequently conflate ecologically rich native forests with commercial monocultures, including oil palm, eucalyptus, and rubber plantations. This creates blind spots in ecological accounting, as a newly planted timber crop can register as intact forest cover.

ForTy addresses this discrepancy by categorising forest typology rather than simple canopy presence. DeepMind trained the system using 1.7 million expert-annotated reference sites worldwide. The dataset incorporates multi-spectral optical imagery from the European Space Agency’s Sentinel-2 satellites alongside auxiliary geospatial features, enabling the system to evaluate structural and phenological variations across diverse biomes.

The resulting model segments landscapes into distinct categories, distinguishing natural forests from commercial tree plantations and agroforestry systems. Operating at a 10-metre ground sampling distance allows the system to resolve field-level boundaries, including smallholder agricultural plots and fragmented forest corridors that coarser satellite analyses obscure.

DeepMind designed the model to support ecological restoration monitoring, carbon accounting, and regulatory compliance. Supply chain auditing frameworks, including the European Union Deforestation Regulation, increasingly require operators to demonstrate that agricultural commodities were not grown on recently deforested land. Traditional canopy metrics often fail to detect when industrial plantations replace cleared primary woodland, because satellite-observed tree cover rapidly rebounds even as biodiversity and carbon stability decline.

By distinguishing planted monocultures from self-sustaining ecological systems globally, ForTy provides an automated tool for conservationists and regulators tracking land-use change over time. DeepMind has released the global forest typology maps and underlying data for public and scientific use.

The primary technical hurdle now shifts toward real-world verification. Will automated optical classification prove robust enough to distinguish degraded native woodlands from complex agroforestry systems across heterogeneous tropical landscapes?